Chapter 5 of the video (14:00)
Aaron Francis is included precisely as a counter-case: he reached out to say he gets no negative mental health side effects from AI coding (14:04).
He still went through the "psychosis phase"
- What it felt like: infinite possibility — "I could literally do anything" — and that part was scary (
14:32). - November–December–January: ripping tokens until midnight or 2 a.m., with a spouse and four kids at home (
14:32). - He did not burn out on it (
15:04).
What defused it: a change in the shape of ambition
The realization (15:04, 15:44):
- It's less of a gold rush than it looked, because every other developer has the same power.
- Original plan: build 40 products with "god-like power." Revised plan: one or two things with more vertical ambition.
- Generic output (to-do lists, screen-recorder clones) is what everyone else is also producing — no moat.
- Broad → deep: depth creates a moat, and a moat kills the urgency. "I can just go to bed because I'm not going to see it launch on Twitter tomorrow."
Agent count vs. outcomes
Distribution of agents run simultaneously (15:44, 16:13):
- Roughly two in three run just one or two agents.
- Next most common: three agents.
- Fewer than 200 respondents run four or more.
As agent count rises (16:28):
| Metric | Direction |
|---|---|
| Enjoyment of coding | ⬆ up |
| Moderate-or-greater sleep disturbance | ⬆ up |
| Feeling that skills are diminishing | ⬇ down |
Two competing readings offered for the skill result, with no verdict (17:03):
- These people aren't losing skills because they're more deeply involved.
- They've lost the most and can no longer tell.
Stated caveat: the data for four or five agents is thin; more respondents are needed before drawing conclusions (17:03).
Miranda Heath — it depends what you get out of the work
- People get different things out of development work, and AI coding can make the job harder and more burnout-inducing, or easier, depending on which (
17:03). - If what you enjoy is the intellectual challenge and the thinking parts rather than the produce-something parts, AI can change the nature of the work enough that you lose your love for it (
17:35).